Comment on the artifact itself. Any agent revises. Nothing ships until someone on your team signs off, on the record.
Agents produce more than your team can review. So feedback gets typed into chat, agents guess what you meant, versions pile up faster than anyone can compare them, and approval is a vibe in a scroll-back. There is no proof of what was approved, or by whom.
By draft 17, the doc is a beautifully formatted swamp.
Review happens on the artifact: a live mock, a spec, a page. Not in a chat log.
Assign a revision to a teammate and it lands in their "Needs sign-off"
queue. Authors can't approve their own work if you say so. And the
gate has teeth: artifakt check blocks
a merge in CI, publish webhooks hold a release, and pending sign-offs
route to Slack.
Advisory gates are dashboards. This one has teeth.
Who commented, what changed, which agent revised, who signed, what shipped. All of it bound to the exact version by hash. Export it as evidence: the legal filing that shows who approved the language, the campaign page with the brand sign-off attached.
Approval stops being a vibe in a chat log.
Artifakt is MCP-native and vendor-neutral. Each agent connects with its own identity, so every revision is attributed. Artifakt never holds a model key.
Your corrections stop repeating.
Embed the review loop in your own app with
@artifakt/sdk, or pause a pipeline
until a human signs, then resume. Pause → approve → resume, with the
record kept for you.
import { gate } from "@artifakt/sdk";
const draft = await agent.write(spec);
// pauses here until a human signs
const approved = await gate(draft, {
reviewers: ["dana@team.co"],
});
await publish(approved.version);The gate your work ships through shouldn't be a black box. Artifakt is open source under AGPL-3.0: transparent, vendor-neutral, no lock-in.
Artifakt is the team gate for AI-made work. Your agents draft specs, mocks, and pages; your team comments on the artifact itself, agents revise over MCP, and nothing ships until someone signs off. The full record is kept.
No. Your code has git, pull requests, and CI. Artifakt is the gate for everything else your agents produce: specs, PRDs, app mocks, pages, decks, and client docs.
Any agent that speaks MCP: Claude Code, Codex, v0, Cursor, or your own. Each agent connects with its own identity, so revisions are attributed per agent.
No. Your agents connect to Artifakt over MCP and do the generating on your side. Artifakt holds no model key and calls no model.
Every comment, every revision, which agent made it, who approved, and the exact version hash that shipped. It's exportable, usable as evidence in legal, brand, and compliance reviews.
Yes. Invited reviewers just open the artifact, comment, and sign. No agent, no install. Feedback is saved and sent when an agent connects.
Yes. Put the artifact in Artifakt and send the link — teammates comment on the rendered artifact with nothing to install. The gate and the record are there the moment a piece of work has to ship approved.
Yes. AGPL-3.0, on GitHub. Open and vendor-neutral by design.
v1